AI
Nemotron Labs: How Open Models Give Enterprises and Nations AI They Can Trust, Control and Customize
NVIDIA highlighted how enterprises and nations are customizing its open Nemotron models to build specialized, trustworthy AI systems.
Key takeaways
- Open models like NVIDIA Nemotron provide complete ownership and control, removing the inspection barriers set by closed models.
- Harvey matched leading closed models on complex legal tasks at at least 10x lower cost per run by post-training Nemotron 3 Ultra.
- H Company's Holotron 3 Nano achieved higher than 76% accuracy on OSWorld-Verified by post-training Nemotron 3 Nano Omni.
- Arcee AI achieved inference costs of roughly 90 cents per million output tokens by post-training Nemotron on the NVIDIA Blackwell platform.
- LangChain achieved top agent accuracy among open models at approximately 10x lower cost per run than leading closed alternatives using Nemotron 3 Ultra.
NVIDIA highlighted how enterprises and nations are customizing its open Nemotron models to build specialized, trustworthy AI systems. By post-training and fine-tuning models like Nemotron 3 Ultra and Nemotron 3 Nano Omni, partners such as Harvey, H Company, and YTL AI Labs are achieving frontier-class accuracy on domain-specific tasks - including legal work, computer-use automation, and local languages - at significantly lower operational costs and without routing proprietary data through third parties.
By the numbers
- 76%
- Accuracy achieved by Holotron 3 Nano on OSWorld-Verified
- 10x
- Lower cost per run achieved by Harvey on legal tasks
- 90 cents
- Inference cost per million output tokens achieved by Arcee AI
- 20x
- Cheaper inference cost achieved by Arcee AI compared to closed models
Turn stories like this into views
Ravenclip finds the AI news, makes the video, and posts it before attention moves on.
Common questions
- What happened with NVIDIA Corporation?
- Open models like NVIDIA Nemotron provide complete ownership and control, removing the inspection barriers set by closed models.
- Where can I read the original report?
- Read the full report at nvidia_blog.
